JSONDecodeError: Extra data: line 1 column 61 (char 60)

[al-yakubovich]

Hi. I am getting the following error:

JSONDecodeError: Extra data: line 1 column 61 (char 60)
Traceback:
File "C:\Users\AppData\Local\anaconda3\envs\py311_test\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 565, in _run_script
    exec(code, module.__dict__)
File "C:\Users\Desktop\GenAI\app\app2_test\interface.py", line 72, in <module>
    decoded_response = decode_response(response)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Desktop\GenAI\app\app2_test\interface.py", line 17, in decode_response
    return json.loads(response)
           ^^^^^^^^^^^^^^^^^^^^
File "C:\Users\AppData\Local\anaconda3\envs\py311_test\Lib\json\__init__.py", line 346, in loads
    return _default_decoder.decode(s)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\\AppData\Local\anaconda3\envs\py311_test\Lib\json\decoder.py", line 340, in decode
    raise JSONDecodeError("Extra data", s, end)

It looks like something wrong with decode_response function.

I changed it to:

def decode_response(response: str) -> dict:
    lines = response.splitlines()
    json_line = lines[0]  
    return json.loads(json_line)

and it started working for simple questions, but it fails for most questions (e.g. plot something)

[al-yakubovich]

Looks like it is all about token limits. When response is too long then system just cut it and response becomes not correct json structure. For example: {'key1': 'long_text_here, 'key2': 'another_long_text_her

[Sharvadze]

Having the same issue, did manage to sort it out? Also, it's pretty slow with 10mb CSV files

[al-yakubovich]

Nope, this tool outputs json structure with all data from csv and it would always hit token limit. The right way to do it is to change prompt and code so it would output pandas/matplotlib code instead and then this code is needed to be converted into pandas df/plot.

[Ngonie-x]

Hey. You're absolutely right. When outputting data as a JSON structure, it's highly likely to hit the token limit. To address this, I'm experimenting with an alternative approach by outputting the data as a data frame formula. For instance, instead of returning a complete string of books with the highest rating, it will return a string like {"table": {"data": "df[['title', 'ratings_count']].head()"}}.To convert this string back to a Python dictionary, we can use json.loads().

Once the string is converted to a dictionary, we can apply it to the actual DataFrame, like this:

df = pd.read_csv(data)

if "table" in response_dict:
        data = response_dict["table"]
        table_df = eval(data["data"])
        st.table(table_df)

The evaluation statement eval() will process the expression, effectively converting it back into a DataFrame object. Finally, the rendered DataFrame will be displayed in Streamlit.